How To Compose Music: Daily Repeated Random Variation

30 October, 2016
A method for composing music: possibly the only method for composing music. Based on your brain's natural learning ability and the concept of a local theory of music.

Who can compose music?

There seem to be many more people who play music than there are people who compose music.

Is there something special about those people who compose the music that other musicians play?

Or is it just that they do something differently, perhaps without even realising it?

The method of composing music described here is one that I developed starting from a fundamental assumption that there is nothing special about those people who compose music.


The prerequisites for this method of composing music are fairly limited, and consist of learning the fundamental concepts of music, such as:

It is useful to be able to play music on some instrument. However it is entirely possible to combine the process of composition with that of practicing the instrument that you are composing on. (The result will be a performance repertoire consisting entirely of your own compositions, and there's nothing necessarily wrong with that.)

The most straightforward instrument for composition is an electronic keyboard, or a piano, because that's the easiest way for one person to play both melody and chord accompaniment simultaneously in real-time.

However the elements of this method can be applied to other instruments, and/or combinations of an instrument and one's own voice.

The Basic Method

The basic method of composing music by repeated random variation is as follows:

  1. Initially improvise a melody, together with a chord sequence. (Typically you would want to use a "well-known" chord sequence, but anything that "sounds nice" will do.) This combination of melody and chord sequence is the starting point.
  2. Perform small random variations of the initial melody. There is no precise rule for what constitutes a "small random variation", and indeed it may be worth experimenting, over a period of time, with different types of variation, to see what works the best.
  3. Repeat the previous step daily.

There is no exact requirement for how much random variation is required in a given session, or within a single day, but I would suggest something like at least 1 hour a day, possibly split over two or three smaller playing sessions within the day.

And as for how long the process takes, I would suggest persisting for at least a few months. (If indeed all musical composition depends on this approach, then the best possible results will come from those who spend hours a day on it for years on end, so it is unlikely for a similar level of success to be achieved within a period of time as short as only a few days or weeks.)

Explanation of How the Method Works

At first it seems that this method is a straightforward prescription for a brute-force search through a space of possible musical items.

However it is more than a search: it is a learning method.

When you play a set of small random variations of a musical item, your brain does more than just learn about the musicality of those specific variantions: it actually teaches itself a local theory of a small sub-space of the set of music possibilities.

It is as if a surveyor starts with altitudes of a small set of points in a small region on a map, and then somehow magically produces a very accurate map of all the points contained within a border perimeter enclosing the initial set of points.

This method works because our brains are optimised for solving this type of learning problem.

As a result of this learning process, your brain will very quickly determine an optimal point of musicality within the region of variation, even though that particular variation was not one of the specific variations that you played earlier.

The daily nature of the schedule plays an important role, because some components of learning occur within our brains when we are not practicing something that we are learning, and some of those components of learning only occur when we are asleep.

Enhancements: Multiple Starting Points and Cross-Fertilisation

The method as described includes an underlying assumption that a strong musical item is to be found somewhere near the starting point, and that this target musical item can be reached after some reasonable amount of time (ie months and not decades). One might be unlucky enough to choose a starting point that isn't actually "near" the kind of strong original music that one is hoping to compose.

To improve one's odds, it would be prudent to work with multiple starting points – but not too many, because if you spread your efforts too thinly, you will not be doing enough learning within the local space of musical possibilities surrounding each individual starting point.


The map metaphor that I gave above is also not entirely accurate, because on a map there is typically little relationship between the neighbourhoods of points that are separated from each other by a larger distance.

Musical items, on the other hand, can be split into components and recombined in ways that don't make sense for map coordinates.

For example, I can take the melody line of item A and combine it with the rhythm of item B, to create a new item C. In this case, what my brain has already learned about variations in a small neigbourhood of A and variations in a small neighbourhood of B is already providing it with information about the likely behaviour within small variations in the neighbourhood of the new item C.

(The map analogy would be that I've visited 30°N 20°W and 15°N 10°E, so now I know all about 30°N 10°E and 15°N 20°W, which doesn't work in practice.)

The possibilities of cross-fertilisation are further reason to keep more than one composition "on the boil" over a given period of time (but still subject to the above-mentioned constraint that spreading yourself too thin will reduce the effectiveness of the learning process for each individual composition).

Deducing This Method From the Mere Existence of Music

The most efficient way to compose new music would be to use a complete theory of music, which would consist of a mathematical formula to calculate the expected musicality of any possible musical item (subject to additional parameters to take into account personal taste, etc etc, but that's not a major factor in this argument).

Such a complete theory of music does not exist. It does not exist, not yet, because science has not yet solved the mystery of what music is. There probably will come a day where science solves the mystery of music, and then a complete theory of music will exist. But right now, there is no complete theory.

A complete theory would provide a formula for calculating musicality, and very likely such a formula could easily be reversed to provide a formula for generating new strong musical items. (Sometimes mathematical formulae are not easily reversed, but in most cases those formulae have been contrived to be hard to reverse, and there is no particular reason why any formula in a scientific theory describing some aspect of nature should happen to fall into this category.)

If follows that even a partial theory would be more efficient than a simple "brute-force" search, or its cousin, the "brute-force hill-climb".

And it further follows that if there is any possible method of generating such partial theories, then that method will generate music of much higher quality than any other possible approach.

So quite possibly every item of strong music that you have ever heard has been composed by a composer who has succeeded in developing, within their brains, a partial theory of music which accurately predicts the musicality of music within some subspace of the space of all possible music, where that subspace happens to contain that musical item.

Indeed there is good reason to suppose that our brains are already optimised to develop such partial theories, for example when learning a physical skill. Learning a physical skill implies the development of a theory of physics that applies to the subset of all physical motions of one's body which are within a small "neighbourhood" of the set of motions required to perform the skillful action being learned. One's brain can learn such a partial theory of physics, in the process of learning a specific physical skill, without ever learning the full theory of Newtonian physics.

If successful composition results from applying the correct method to invoke brain capabilities that every person alive has, then there is indeed nothing special about people who compose music, any more than there is anything special about people who can ride a bicycle.

(It was following this same chain of reasoning in reverse that lead me to develop this method in the first place: ie if we assume that there is nothing special about people who compose music, and given the lack of any scientific understanding of the nature of music, then by whatever method can it be possible for anyone to compose music at all?)


The only way to really verify the effectiveness of this method is to go out and do it.

I can present the results of my own efforts, at My compositions are not anything world famous, but they are somewhat better than I ever expected myself to be able to achieve.

Sufficiency and Necessity

I do make the strong claim that not only does this method work, but that almost any successful musical composition is the result of some variation on this method, ie that it is not only sufficient for the composition of music, but also necessary.

To verify this claim of necessity, the following steps are required:

Despite the apparent simplicity of this research plan, I suspect that achieving the required levels of accurate recall and honest recounting of those accurately recalled events might be easier said than done.

An important aspect of the method of repeated variation is that it involves learning processes which one is not consciously aware of, so the composition of any particular musical item may be the result of any and all past intentional musical performances or improvisations by the composer, without the composer being consciously aware of the connection in each case between a particular set of performances and the development of the final composition that may have arisen from them.

(What I can predict, and this might be easier to verify, is that if a musician systematically avoids any type of variation, and only ever plays well-known musical items in a manner replicating well-known performances and recordings, then that musician will be much less likely to ever compose strong original music.)